How do I use SentryDock with ChatGPT?

A first task for ChatGPT

Compare the last 72 hours of reported LNG terminal outages by facility. Return facility, country, event, publication time and source URL. Mark conflicting reports and do not infer an outage from an empty search.

Ask the agent to show the tools it called and the returned sources. Review the links and dates before using the note. This is an example prompt, not a recorded product result.

Check access before searching

Ask “Use account.overview to show my plan, active monitors, delivery settings and latest alerts.” The returned account should be yours. News and monitoring need a paid plan; overview and usage can be inspected before payment. If the tool returns plan_required, open its pricing link and choose a plan yourself.

A saved connector only shows configuration. A successful account.overview response establishes account access; a source-linked news result establishes that search worked. Start with those two read-only calls.

Search, then write a source-linked note

Use news.search for a topic and a time window, news.latest for the newest indexed items, or news.mine and alerts.list for your own monitored hits. Search accepts query, hours_ago and limit; hours_ago can cover up to 720 hours. Preserve the article URL, publication time when available, and original title.

SentryDock searches its existing index. It does not crawl every source on demand. An empty result means no matching indexed item was returned. Broaden the time window, try a named company or location, and inspect the source coverage before concluding anything about an event.

Turn a one-off question into ongoing monitoring

Ask the agent to inspect monitors.list before proposing a new monitor. Specify the company or topic, source URLs, relevance rules, cadence and delivery destination. Review the concrete proposal, then authorise monitors.create. The default interval is hourly; realtime availability and source limits depend on your plan.

After creation, read the saved monitor and its schedule. Use monitors.run for an authorised test, monitors.runs for status, and monitors.results or alerts.list for output. A saved monitor is not a completed run, and a completed run is not proof a notification arrived.

Example: a monitor that keeps working after the chat

“Inspect my existing monitors. Propose an hourly monitor for the company names and official newsroom URLs I provide. Include material regulatory actions, operational disruptions and earnings announcements. Exclude repeated commentary. Show me the sources, cadence and current delivery before saving.”

Once approved, SentryDock runs the saved monitor independently of the chat. The assistant does not receive a live push simply because MCP is connected. Read alerts on demand or connect a signed webhook to a reachable HTTPS endpoint you control.

Troubleshooting the connection

• No tools: enable the connector for this session, reload MCP or start a new chat, and check the workspace allowlist.
• Authentication error: reconnect with OAuth, or check that your API key is active in Settings → Agents. Keep real keys out of shared configuration and chat.
• Wrong account: disconnect and authenticate with the intended SentryDock account before reading or changing monitors.
• plan_required: use the pricing URL in the response. Connecting an agent does not activate a paid plan.
• Empty or stale news: check the requested time window, named sources and monitor run history; a keyword match is not evidence of full coverage.
• A write is rejected: inspect the current schema and confirmation requirements; do not retry an uncertain monitor creation without checking for a duplicate.

Frequently asked questions

Does this install a marketplace app? These instructions use a custom remote MCP connection. A public marketplace listing is a separate distribution route.

Does installing a skill connect my account? No. The skill gives the agent a workflow; MCP OAuth or a securely configured API key provides access.

Can the agent change my billing? The gateway does not support changing billing, team invites, predictions or deleting your user.

How do I stop access? Remove the connector from the client. For API-key access, revoke the key in Settings → Agents. Pause or delete monitors separately if you want monitoring itself to stop.

Client instructions and related guides

Setup checked against OpenAI documentation on 3 October 2026. Menus and workspace policy can change; these steps document the setup method rather than an authenticated test in every client.

Install the SentryDock skill · Company watch · Market briefing · Signed agent callbacks